It is shitty for infrastructure. 99.9% server uptime means your site might be down for about 44 minutes per month, through no fault of your own--i.e. that does not even take into account scheduled downtime, application problems, DDOS attacks, etc.

really, you don't count scheduled downtime and DDOS attacks in your uptime?

Might as well stop counting sysadmin screwups and hardware failures, at that point. 100% uptime! except when we're down.

I do count scheduled downtime and DDOS attacks in my uptime, which is why I wouldn't tolerate 44 additional minutes of downtime from my infrastructure provider.

yeah, my argument is that scheduled downtime (scheduled by the infrastructure provider) and ddos attacks (against the infrastructure provider or other customers of the infrastructure provider) should count against the infrastructure provider's uptime.

The current balance of power is tipped, right now, so hard in favor of the attackers that if anyone wants to take /you/ out, well, it's pretty difficult (read: expensive) to stop them, and if you are some $20/month customer, your upstream is probably going to just finish the job and cut you off... but if the guy next to you gets attacked and you are collateral damage? You should count that against your service provider. There are usually things we can do (as service providers) to limit collateral damage, even in the cases where we are unable to protect the target. If nothing else, providers who tend to tolerate sites that get attacked often, well, they suffer collateral damage from attacks, generally speaking, more often than sites that are less tolerant. It's sad, really 'cause sometimes the targets are not doing anything really wrong, but you've gotta protect your network.

There is some advantage of scale here, too... The DoS that is hardest to fight is the pipe-filling attack. If the attacker can fill your upstream port(s), then there isn't really much you can do, besides blocking the target at your upstream, (I mean, assuming the source is random, as it often is.) The larger your upstreams are, the more you can absorb before you have a choice between cutting off the target customer and having all your customers cut off.

Do you consider a 45-minute outage once a month to be acceptable? Maybe for the price, it is.

Well, if we are being pragmatic it sounds totally acceptable for any kind of business.

Except maybe if those outages hit them at the worst possible moments of the month (just before a major announcement or something).

Now, managers would froth at the mouth hearing 45 mins/month is acceptable. But those managers are idiots.

I haven't seen any major site that didn't have those (and worse) outages.

5-6 hours a year? Heck tons of sites had those. AWS had them (and thus Heroku had them), GAE had them, Twitter had them, Tumblr (well they had one-two orders of magnitude more), etc.

45 mins a month represent lots of lost revenue to you? Really? If you're not Google or something, then what exactly do you do?

Shouldn't that be like 0.1% of your revenue? And that IF you sell online and IF we assume customers aren't gonna come back. Heck, you lose more because of random fluctuations in buyer preferences than that.

Depends how you look at it. A minute and a half a day? Doesn't sound too bad when you say it that way. A 3-hour outage every 4 months? My cable company can barely do that.

The problem is outages tend to be for prolonged periods and not spread evenly across the month. A minute may not seem like a huge deal, but 3 consecutive hours is really going to piss some people off. If you run a business, it's completely irresponsible to provide such a low level of service to your customers.

0.1% downtime is about 45 minutes per month. Depending on how it's distributed, that can be extremely annoying.

Well, these days it is. You want five nines (99.999%) on mission-critical applications.

Five nines will cost you a lot of money.

it's usually achieved through luck, not engineering.